Short-term Traffic Speed Prediction of Urban Road Networks using the Integration of Spatio-Temporal Graph Convolutional Networks and Convolutional Neural Networks

Keywords: Spatio-Temporal Graph Convolutional Network, Convolutional Neural Network, Short-term Traffic Speed Prediction
Abstract Type: Paper Abstract

Abstract

The significant growth of large cities leads to the importance of road infrastructure. A deeper understanding of road networks can improve the quality of service and reduce fuel consumption. It can also help to optimize travel plans by knowing the condition of the road network in advance. However, the road network has a very complex environment and is constantly affected by various factors and state changes. It is challenging to understand and predict the flow and condition of these road networks. Recent research has made efforts to identify and predict the flow and state of road networks using deep learning. Extracting significant features from various factors can significantly improve the model for prediction accuracy. This study conducted a short-term prediction of the traffic speed of road networks. Experimental data come from Mobileye sensors attached to taxis in Daegu city, Korea. These sensors can simultaneously reflect the environment of the road network and the driver's intention. Based on these data, this study integrated a Convolutional Neural Network (CNN) with Spatio-Temporal Graph Convolutional Networks (STGCN). The experimental results demonstrated that the integration of STGCN and CNN performed better than the STGCN and CNN model, respectively. The findings of this study can improve the development of short-term prediction models for traffic speed, resulting in the enhancement of road network management.